First Article Inspection: What Importers Should Check Before the Run

A first article inspection is a one-time, complete check of the first unit produced with the actual production tooling, against every characteristic the order defines, carried out by someone other than the person who made it, before the rest of the run is authorised. The aerospace standard that defines it, IAQG/SAE 9102, states the purpose plainly: to “validate that the product realization processes are capable of producing parts and assemblies that meet engineering, design requirements.”

Almost everything written about it assumes you are buying a machined titanium bracket against a dimensioned CAD drawing, and that you own a coordinate measuring machine. If you are buying 8,000 melamine bowls or a container of LED string lights out of Yiwu, you have neither. That does not mean the control does not apply to you. It means the paperwork does not, and the underlying idea does — and nobody has written down which parts survive the translation.

On this page

• What a first article inspection actually is
• What the standard requires, and what it pointedly does not
• The Yiwu translation: FAI when there is no drawing
• FAI vs PPAP vs AQL sampling
• What to check on the first unit
• What to put on the PO so the first unit actually gets checked
• When the first article fails
• So is it worth doing on a Yiwu order?
• Frequently asked questions

What a first article inspection actually is

The word “first” is doing precise work. It does not mean the sample the supplier couriered you before you paid the deposit, and it does not mean the first carton off the pile at the end. It means the first unit that came out of the same tooling, on the same line, with the same material and the same operators that the remaining thousands will use. That is the whole point: you are testing the setup, not the object.

IAQG’s own guidance lists what the exercise is supposed to buy you — “reduce future escapes, risks, and total costs,” and “identify non-capable production realization processes, initiate and validate corrective actions.” Read that second one again, because it is the one importers keep missing. The output of an FAI is not really a good unit. It is information about whether the process can keep producing good units after you have stopped watching.

This is why the timing is not negotiable. A first article checked after the run is finished tells you what went wrong; a first article checked before the run is released tells you what is about to go wrong, while changing it still costs a morning instead of a container.

It also explains why the courier sample keeps disappointing people. In a Yiwu order that sample was often made by hand, pulled from booth display stock, or produced by a different workshop than the one that will run your quantity. It is evidence about what the supplier can make at their best — not about what the tooling produces at volume, which is the only question that determines what arrives in your container.

What the standard requires, and what it pointedly does not

The reference document is 9102, published by SAE as AS9102 and maintained by the International Aerospace Quality Group. It is worth reading the actual requirements rather than the vendor summaries, because three of them transfer perfectly to a consumer goods order and one of them is routinely misrepresented.

The first is independence, and the standard is blunt about it. Asked what “independent” means, IAQG answers: “The person that verifies the characteristic for the First Article cannot be the same person that generated the characteristic. Self-inspection is not permitted (ie. operator self-verification). Also, the equipment used to verify the characteristic needs to be different from the equipment used to produce the characteristic.”

That second sentence is the one to carry into a workshop. If the same jig that pressed the part is used to check the part, the check cannot detect a wrong jig — which is the single most common cause of a whole run being uniformly, confidently out of spec.

The second requirement is traceability of every characteristic. The standard asks that “every design characteristic requirement is accounted for, uniquely identified, and has inspection results traceable to each unique identifier.” Note what that sentence demands: identification and traceability. It does not demand a particular document.

The ballooned-drawing myth. Ask IAQG directly whether a ballooned drawing is required in the report and the answer is no: the ballooned drawing “is often used to show verification that ‘every design characteristic requirement is accounted for’… However, there is no requirement in 9102 for a ballooned drawing to be used, so there no requirement for it to be part of the supporting documentation.” The standard’s own position is that “Standards provide requirements and cannot provide methods. You may use any technique that provides traceability from the engineering to the FAI report.” This is the clause that lets the concept travel to an order that never had a CAD file.

The third is what Revision C changed, and it should alter how you read every inspection report you are sent. IAQG’s summary of changes from Revision B to Revision C, dated 11 July 2023, records under section 1.2 Purpose: “Added clarification FAI is not a product acceptance document.” A first article report is evidence about a process — not a receipt, not a sign-off on the goods, and not a waiver of your right to reject the shipment later. Suppliers and occasionally inspection agencies present a signed report as though it closes the quality conversation. Under the current revision it explicitly does not.

Revision C tightened three other things worth knowing:

• 1.3 Application — “Removed unique single run production order exemption.” The carve-out that let a one-off production order skip FAI is gone.
• 1.3 Application — “Added clarification that each item shall have a separate FAI.” One report does not cover a mixed order.
• 4.1 First Article Planning — the organisation “shall have a documented process to plan for FAI,” rather than improvising one per job.

The Yiwu translation: FAI when there is no drawing

Here is the situation the standards literature does not describe. Your supplier is a booth in one of the market districts — a trading operation that holds the customer relationship and places the order with a workshop somewhere in the surrounding counties, which you may never see. There is no engineering drawing. There is a physical sample you approved, some photographs, a Pantone reference if you were organised, and a WeChat thread. The reference is not a drawing; it is that approved sample, and everything depends on pinning the object down before production starts, because it is the only definition of “correct” in existence.

How the reference sample gets fixed in practice. In the market districts this object is usually called the golden sample, and it is worth using that term because it is the one a booth will recognise instantly. The workable arrangement is three identical units from the approved sample batch, not one — each tagged with the order number, the date and both signatures:

• One stays with the booth.
• One goes to the workshop that will actually produce.
• One is held by whoever inspects on your behalf, or shipped to you.

The tag matters more than the seal. A bag sealed with tape and no writing on it is indistinguishable from any other bag in six weeks, and the argument you are preventing is precisely the one where two people hold different objects and both are sincere.

Where this goes wrong is nearly always the middle copy: the booth keeps its sample and neglects to send one onward, so the people actually making your goods work from a verbal description relayed through a trading company. Asking who physically holds the workshop’s copy, and getting a photograph of it with the tag legible, is the cheapest quality control available on a Yiwu order.

What the booth will sign, and what it will not. This is worth knowing before you draft anything, because asking for the wrong document produces a polite refusal and a worse relationship.

That last row is not dishonesty, it is the business model, and the practical consequence is specific: if the booth will not name the workshop, your independent verification has to happen where the goods physically arrive — the consolidation warehouse — rather than at the point of manufacture. You lose the ability to catch a bad setup on day one. You keep the ability to catch it before the container is sealed. That trade-off is worth understanding when you decide whether you are dealing with a trading company or a factory, because the answer changes what verification is even physically possible.

FAI vs PPAP vs AQL sampling

These three get used interchangeably in supplier emails and they are not interchangeable. They answer different questions at different moments, and the confusion is expensive because a buyer who thinks he has bought one control has often bought another.

PPAP is the automotive equivalent and a heavier instrument: eighteen documented elements summarised by a Part Submission Warrant, at one of five submission levels governing how much evidence is actually sent rather than retained at the supplier. Level 1 is the warrant alone; Level 3 — warrant, samples and complete supporting data — is what most customers ask for by default; Level 5 keeps the full package at the supplier’s plant for review there. No Yiwu trading booth runs PPAP, and asking for it signals a requirement copied from a template rather than thought about.

The AQL confusion is the costly one. An acceptance quality limit is a lot-acceptance rule, not a defect allowance, and ISO 2859-1 says so in a caution printed in the standard: “The designation of an AQL shall not imply that the supplier has the right knowingly to supply any nonconforming item.” The same clause separates the two ideas buyers merge — the AQL “is a parameter of the sampling scheme and should not be confused with the process average that describes the operating level of the manufacturing process. It is expected that the process average will be better than the AQL.” Two further clauses are useful leverage:

• The AQL “shall be designated in the contract or by… the responsible authority” — your grounds for putting it on the purchase order instead of leaving it to whatever the inspection agency defaults to.
• Even on an accepted lot, “the right is reserved to not accept any item found nonconforming during inspection.” A passing lot does not oblige you to keep the defective pieces the sampling surfaced.

For how the sampling levels and the inspection stages fit together on a real order, the site’s guide to China QC inspection stages and AQL standards covers the plan itself; this page deliberately stops at the boundary rather than reproducing it, and the Yiwu-specific AQL walkthrough goes further into how the levels get chosen.

What to check on the first unit

What to check on the first unit, in order. Work down this list before anything else is produced; each item is chosen because it is the point where a Yiwu order most often diverges from what was agreed, and because a non-specialist can apply it without instruments:

1. Identity. Is this unit from the production tooling, or is it the courier sample photographed again? Ask for it beside the tagged reference in the same frame.
2. Colour, against the tag, in daylight. Booth lighting flatters. This is the single most disputed characteristic on decorative goods.
3. Print and logo. Position, size, orientation and spelling — logos get re-typeset by the workshop more often than buyers expect.
4. Material and weight. Weigh it. A cost-down substitution in resin, fabric or metal gauge shows up as a weight change before it shows up anywhere else.
5. Function. Cycle every moving part, switch, zip or clasp the number of times the retail buyer will in the first week.
6. The hidden characteristic. Anything inside a wall, under a coating or behind a glued seam has to be seen now — after assembly it is unverifiable until a customer finds it.
7. Retail packaging. Check it as a unit with the product inside, not as a flat sample, and drop it.

What to put on the PO so the first unit actually gets checked

An agreement made in a WeChat voice message does not survive a production schedule. The clauses below are short on purpose — a five-line requirement gets read and signed, and a long form gets a signature that means nothing.

• Name the reference and its custody. “Production must match the sealed reference sample tagged [order number / date], of which three tagged copies exist: buyer, seller, producing workshop.” Without the custody clause you have a reference nobody can produce.
• Name the hold, and bound it. “Seller shall photograph the first 3 finished units against the tagged reference and shall not continue production until buyer confirms, within 24 hours of receipt.” The bound is what makes it acceptable to the supplier; an open-ended hold is what gets quietly ignored.
• Name the characteristics. Three to six from the checklist above, in writing, each with a pass condition stated so a non-specialist can apply it without judgement calls.
• Name who checks. Per the standard’s independence rule, not the operator who produced it. In a Yiwu chain this means your agent, a third-party inspector, or at minimum a named person at the booth who is not the person running the line.
• Name the AQL and its edition. ISO 2859-1 requires the AQL to be designated in the contract. State the inspection level and the acceptance quality limits for major and minor defects, and state which edition’s tables apply.
• State what a passed first article does not do. One sentence: approval of the first article authorises production only and does not constitute acceptance of the shipment. This mirrors what Revision C added to the standard and it closes the argument before it starts.

What we check, and where we stop. Our published stages are triggered by production percentage — during-production at 20–50% produced, pre-shipment at 100% produced and 80% packed, loading supervision on stuffing day — and the earliest of those still sits well past the moment this article is about. Our defaults are an AQL plan, not characteristic-by-characteristic verification: critical zero allowed, major AQL 2.5, minor AQL 4.0, against ANSI/ASQ Z1.4 (ISO 2859-1) tables, adjustable on request, reported as PDF within 24 hours with photos, function-test video and measurement data.

So: no AS9102 FAIR, no metrology lab, no quality-system certification, and no first-article stage in the published service either. Wanting the first unit held means writing the clauses above onto the PO and booking it as a visit, not assuming it into an AQL package.

What we hold that a pure inspection company does not is money. A third-party inspector reports a problem and leaves; on orders we place, the final 70% payment sits with us, and a failed check means a “Do Not Ship” alert with that balance withheld until the goods are reworked and a re-inspection passes. That is what makes a hold enforceable where the booth’s incentive is to keep the line moving — a signed clause with the balance already paid is a request, not a hold.

Two limits stated rather than buried. The standalone rate published on our quality-control page is flat and daily, given there as e.g. $199 per man-day — an illustration, not a quote, since the real driver is how many separate locations the goods must be seen in, and a booth that will not name its workshop forces that count up. On the red-envelope problem the counter is procedural rather than a promise: inspector rotation so nobody is sent twice, GPS and body-cam recording, and a reward for factories reporting a bribery attempt.

Where a first-article hold is the right control, the role is to be the independent pair of eyes the standard requires, documenting what was seen against your tagged reference — not aerospace paperwork that would mean nothing coming from a trading chain. Two things we will not put a number on, because neither has been verified for your order: spec range is whatever the booths already tool for, so anything outside it becomes a custom job with different economics; and minimum order quantity moves with carton quantities rather than round numbers, which is why an MOQ quoted without your size and colour mix is meaningless.

When the first article fails

A failed first article is the system working: it has cost you one unit and a delay, at the only moment when the alternative was cheap. The standard’s handling is documented rather than emotional. The nonconforming characteristic is recorded on Form 3 with the nonconformance document reference number in field 11, and the organisation then “implements corrective actions and performs a partial FAI for all affected characteristics on the next production run after implementation of the corrective action.”

Note the word partial. You re-verify what changed and what the change could have affected — cheaper, and the reason a supplier will agree to it at all. The standard lists the triggers for that partial re-check:

• A change that can potentially affect fit, form or function.
• A lapse in production of two or more years.
• Moving production to a different location or supplier — where IAQG’s answer is that “you or your new supplier must perform a partial FAI covering the processes/characteristics moved.”

That last trigger is the practical warning for this market. A booth that subcontracts your repeat order to a different workshop than last season has, in the standard’s terms, moved production — and the reference sample your first run was verified against sits with a workshop that is no longer making your goods. Repeat orders fail this way far more often than first orders, because everyone assumes the verification carried over.

One more constraint worth designing around: characteristics that cannot be measured on the finished product “shall be verified during the manufacturing process, as long as they are not affected by subsequent operations.” If it matters what is inside the item, inside the wall, or under the coating, it has to be seen before the item is closed — once the box is glued and the shrink-wrap is on, that characteristic is gone until a customer finds it. If something does go wrong after the goods ship, the recovery path is a separate discipline; the site covers what leverage you retain over defective goods once payment has moved.

So is it worth doing on a Yiwu order?

Honestly, not always, and the trade-off is legible enough to decide case by case rather than by policy.

It is worth the hold when the order has a custom element — your logo, your colour, your packaging, your retail box — because those are the characteristics with no market default to fall back on and the ones a workshop will interpret rather than replicate. It is worth it when the item has a function that can fail, when the value concentrated in one container is large enough that a uniform defect is a business event rather than an annoyance, and on the first run with any new workshop, which includes a repeat order the booth has quietly moved.

It is not worth the friction on a stock item you are buying as-is from booth inventory, in small quantity, with no customisation and a low unit value. There the market standard is itself the reference, an AQL check at the pre-shipment stage is proportionate, and inserting a formal first-article hold buys you a delay and a slightly puzzled supplier. The sampling strategy that precedes all of this is where that call actually gets made.

There is no order value or unit count above which a first article becomes mandatory, and any supplier or agent who quotes you one is describing their own booking policy rather than a rule. The standard sets no threshold at all; the trigger is the presence of a characteristic somebody has to interpret, which is why a 500-piece run of custom-printed boxes can justify the hold while 10,000 pieces of stock tableware does not. Specify the characteristics you want held — dimensions, colour against a named reference, print position, material, function — rather than a quantity, because the characteristic list is what an inspector can actually work from.

What does not change with order size is the part of the standard that costs nothing: somebody other than the person who made it should look at the first one, against a written definition of correct, before the rest get made. Aerospace wrote that down because parts fall out of the sky. It holds up in a market-district workshop for the same reason it holds up in a machine shop — the first unit is the only cheap place to find out that everyone understood the order differently.

Frequently asked questions